![]() ![]() To echo others who have posted there are numerous factors that need to be considered when looking at virtual machine performance issues. Having a VM which attempts to consume 100% of available vCPUs means that it is always in contention for resources with every other VM on the host, as well as with the host itself. Giving it more vCPUs than the minimum it needs just leads to wasted computation cycles and VMs waiting for free resources unnecessarily. Which is among the reasons why you want to pare VM vCPU allocation down whenever possible. Whereas a party just 25% smaller (32 diners) would be able to be seated immediately, even with 4 very active parties of 2 diners. And that happens every time that 40 diner party wants to use the restaurant. The problem with having a party of 40 diners is that in order for it to be seated, it has to wait until the restaurant is completely empty. That can be 40 parties (VMs) of 1 diners (vCPU), 5 parties of 8 diners, 1 party of 40 diners, or any combination which does not exceed the available seat count of 40. Yours, for example, can seat 40 diners simultaneously. Think of your hypervisor as a restaurant, diners as vCPUs, and parties as VMs. Second, the best metaphor I have heard for explaining why higher vCPU count can end up decreasing performance is this: First, you are correct, you are presented with 40 vCPUs because hyperthreading presents two logical processors for each physical core.
0 Comments
Leave a Reply. |